Assessing Machine Learning Models on Temporal and Multi‐Sensor Data for Mapping Flooded Areas DOI Open Access
Rogério Galante Negri, Fernando B. Da Costa, Bruna Ferreira

и другие.

Transactions in GIS, Год журнала: 2025, Номер 29(2)

Опубликована: Март 17, 2025

ABSTRACT Natural disasters, particularly floods, are escalating in frequency and intensity, disproportionately impacting economically disadvantaged populations leading to substantial economic losses. This study leverages temporal multi‐sensor data from Synthetic Aperture Radar (SAR) multispectral sensors on Sentinel satellites evaluate a range of supervised semi‐supervised machine learning (ML) models. These models, combined with feature extraction selection techniques, effectively process large datasets map flood‐affected areas. Case studies Brazil Mozambique demonstrate the efficacy methods. The Support Vector Machine (SVM) an RBF kernel, despite achieving high kappa values, tended overestimate flood extents. In contrast, Classification Regression Trees (CART) Cluster Labeling (CL) methods exhibited superior performance both qualitatively quantitatively. Gaussian Mixture Model (GMM), however, showed sensitivity input was least effective among tested. analysis highlights critical need for careful ML models preprocessing techniques mapping, facilitating rapid, data‐driven decision‐making processes.

Язык: Английский

Assessing Machine Learning Models on Temporal and Multi‐Sensor Data for Mapping Flooded Areas DOI Open Access
Rogério Galante Negri, Fernando B. Da Costa, Bruna Ferreira

и другие.

Transactions in GIS, Год журнала: 2025, Номер 29(2)

Опубликована: Март 17, 2025

ABSTRACT Natural disasters, particularly floods, are escalating in frequency and intensity, disproportionately impacting economically disadvantaged populations leading to substantial economic losses. This study leverages temporal multi‐sensor data from Synthetic Aperture Radar (SAR) multispectral sensors on Sentinel satellites evaluate a range of supervised semi‐supervised machine learning (ML) models. These models, combined with feature extraction selection techniques, effectively process large datasets map flood‐affected areas. Case studies Brazil Mozambique demonstrate the efficacy methods. The Support Vector Machine (SVM) an RBF kernel, despite achieving high kappa values, tended overestimate flood extents. In contrast, Classification Regression Trees (CART) Cluster Labeling (CL) methods exhibited superior performance both qualitatively quantitatively. Gaussian Mixture Model (GMM), however, showed sensitivity input was least effective among tested. analysis highlights critical need for careful ML models preprocessing techniques mapping, facilitating rapid, data‐driven decision‐making processes.

Язык: Английский

Процитировано

0